In this study, we focused on skipped exons (SE), the dominant type of splicing event in the brain, including in the prefrontal cortex [47, 48]. A predictive model was built for each SE to determine the extent that genetic variants could explain the splicing outcome. We predicted the genetically determined inclusion levels (\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${{{\hat{\mathrm \Psi }}}}$$\end{document}Ψ^) for a total of 41,109 SE events annotated in Gencode using the RNA-seq data and imputed genotypes from the CommonMind Consortium (CMC) [21]. The overall workflow is depicted in Fig. 1A. After filtering for the number of junction reads (>10), number of samples (>100), and PSI variability (IQR > 10%), there were 6284 SE events remaining for analysis. For each SE, we used a semi-supervised method to select the SNVs that were most explanatory of the PSI variability. Then we applied the elastic net algorithm to determine the marker SNVs for PSI prediction. Although we initiated the modeling using more relaxed criteria, we found that all of the final selected variants had MAF ≥ 5%, and 90% of them had imputation scores over 0.8, indicating that our approach converged on higher confidence SNVs.
